Certezas e incertidumbres sobre los efectos del cambio climático en el Mediterráneo

From Jonathan Brophy and Daniel Lowd. 2017 AAAI Workshop on Artificial Intelligence and Cyber-Security (AICS). San Francisco, CA.

We have shown the benefits of using a model that can leverage the

underlying connections present between the data in Soundcloud. With the aid of an independent model to mark clear instances of spam and provide a starting point of information, our relational model can work to identify the perhaps intentionally obfuscated comments missed by the independent classifier.

We have seen that adding more rules to a relational classifier can increase its performance, but adding too many can cause a bottleneck in computation time. It is not hard to see the benefits of using a relational model, where implementations like PSL make it very easy to express simple rules that can capture complex relationships throughout the data.

One final aspect to note about this approach is the relationship between the independent and relational models. Since the predictions from the independent model are fed into the relational model, any performance improvement in the independent model will most likely translate to improved performance for the relational model as well. Thus, more in depth natural language processing (NLP) features could be engineered for the independent model, but these are not explicitly necessary to show the relative improvements of the relational model.

The next step of this work would involve testing this model on a different, but similar domain to see if these results can be replicated. YouTube.com would make an excellent choice, as its popularity certainly attracts many spammers, and its social network structure is similar to that of SoundClouds’. Tracks could be

replaced by videos, since users post comments to other user’s videos the same way users post comments to other user’s tracks. All the other rules can essentially stay the same.

There is also the opportunity to learn weights in one domain, and then test their effectiveness on another domain. Also, more work needs to be done on characterizing the practical size of data instances that can be jointly labeled at one time, and how this characterization changes as the number of rules increase or decrease.

One segment of the data that was not used involved spam warnings and spam reports. The ability of one user to flag other users is a common feature in most social networks, and this information can lead to clues about who the spammers are, as well as the credibility of users doing the flagging, as in Fakhraei et al. (2015).

The applications for this kind of model are not bound to social networks. Any type of data that houses underlying relations can benefit from this

methodology, and it is exciting to see what other domains relational machine learning will impact.

REFERENCES CITED

Alvarez-Hamelin, J. I., Dall’Asta, L., Barrat, A., & Vespignani, A. (2005). Large scale networks fingerprinting and visualization using the k-core

decomposition. In Advances in neural information processing systems (pp. 41–50).

Bach, S. H., Broecheler, M., Huang, B., & Getoor, L. (2015). Hinge-loss markov random fields and probabilistic soft logic. arXiv preprint arXiv:1505.04406 . Bach, S. H., Huang, B., London, B., & Getoor, L. (2013). Hinge-loss Markov

random fields: Convex inference for structured prediction. In Uncertainty in Artificial Intelligence (UAI).

Breiman, L. (2001). Random forests. Machine Learning, 45 , 5-32.

Davis, J., & Goadrich, M. (2006). The relationship between precision-recall and roc curves. In Proceedings of the 23rd international conference on machine learning (pp. 233–240). New York, NY, USA: ACM. Retrieved from http://doi.acm.org/10.1145/1143844.1143874 doi:

10.1145/1143844.1143874

Fakhraei, S., Foulds, J., Shashanka, M., & Getoor, L. (2015). Collective spammer detection in evolving multi-relational social networks. In Proceedings of the 21th acm sigkdd international conference on knowledge discovery and data mining (pp. 1769–1778).

Kimmig, A., Bach, S. H., Broecheler, M., Huang, B., & Getoor, L. (2012). A short introduction to probabilistic soft logic. In NIPS Workshop on Probabilistic Programming: Foundations and Applications.

Laorden, C., Sanz, B., Santos, I., Gal´an-Garc´ıa, P., & Bringas, P. G. (2011). Collective classification for spam filtering. In ´A. Herrero & E. Corchado (Eds.), Computational intelligence in security for information systems: 4th international conference, cisis 2011, held at iwann 2011,

torremolinos-m´alaga, spain, june 8-10, 2011. proceedings (pp. 1–8). Berlin, Heidelberg: Springer Berlin Heidelberg. Retrieved from

http://dx.doi.org/10.1007/978-3-642-21323-6 1 doi: 10.1007/978-3-642-21323-6 1

Lee, K., Caverlee, J., Kamath, K. Y., & Cheng, Z. (2012). Detecting collective attention spam. In Proceedings of the 2nd joint wicow/airweb workshop on web quality (pp. 48–55).

Liu, L., Lu, Y., Luo, Y., Zhang, R., Itti, L., & Lu, J. (2016). Detecting” smart” spammers on social network: A topic model approach. arXiv preprint arXiv:1604.08504 .

LOGIC, L., Garrido, A., & Garrido, A. (2010). Fuzzy boolean algebras and lukasiewicz logic..

Low, Y., Gonzalez, J., Kyrola, A., Bickson, D., Guestrin, C., & Hellerstein, J. M. (2010). Graphlab: A new framework for parallel machine learning. In Uai. Newman, M. E., Strogatz, S. H., & Watts, D. J. (2001). Random graphs with

arbitrary degree distributions and their applications. Physical review E , 64 (2), 026118.

Page, L., Brin, S., Motwani, R., & Winograd, T. (1999). The pagerank citation ranking: bringing order to the web.

Pearl, J. (2014). Probabilistic reasoning in intelligent systems: networks of plausible inference. Morgan Kaufmann.

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., . . . Duchesnay, E. (2011). Scikit-learn: Machine learning in python. Journal of Machine Learning Research, 12 , 2825-2830.

Quinlan, J. R. (1986). Induction of decision trees. Machine Learning, 1 , 81-106. Rayana, S., & Akoglu, L. (2015). Collective opinion spam detection: Bridging

review networks and metadata. In Proceedings of the 21th acm sigkdd international conference on knowledge discovery and data mining (pp. 985–994).

Richardson, M., & Domingos, P. M. (2006). Markov logic networks. Machine Learning, 62 , 107-136.

Schank, T., & Wagner, D. (2005). Finding, counting and listing all triangles in large graphs, an experimental study. In International workshop on experimental and efficient algorithms (pp. 606–609).

Sen, P., Namata, G., Bilgic, M., Getoor, L., Galligher, B., & Eliassi-Rad, T. (2008). Collective classification in network data. AI magazine, 29 (3), 93.

Wang, D., & Pu, C. (2015, Aug). Bean: A behavior analysis approach of url spam filtering in twitter. In 2015 ieee international conference on information reuse and integration (p. 403-410). doi: 10.1109/IRI.2015.69

Zhu, Y., Wang, X., Zhong, E., Liu, N. N., Li, H., & Yang, Q. (2012). Discovering spammers in social networks. In Aaai.